# Local Models block

> Local model settings with a GPU server cutaway whose card LEDs show model memory

Size open weight models on your own GPU server: an isometric server whose card LEDs show memory in use, model rows, an inspector, and a header to review pending changes, fit, copy, download or restart. ReUI Frame, Alert; shadcn Slider, Popover.

- Type: Pro block
- Category: [AI Usage](https://reui.io/blocks/ai-agents/ai-usage.md)
- Page: https://reui.io/blocks/ai-agents/ai-usage/ai-usage-4
- MCP: `get_block("ai-usage-4")`
- Built with: Alert, Alert Dialog, Avatar, Badge, Button, Button Group, Dropdown Menu, Field, Frame, Icon Tile, Item, Popover, Progress, Select, Slider, Spinner, Switch, Toggle Group, Tooltip
- Installs with a ReUI Pro or Ultimate license; setup: https://reui.io/docs/license-setup

## Best for

Settings for open weight models on a workspace GPU server, sized by precision, context and parallel requests

## Agent notes

- `MODELS`, `APPLIED` and `SERVER` in data.tsx hold the cards and running config; every memory figure and LED reads `memoryOf` and `loadsOf`
- `handleApply` and `handleRestart` only wait and fire a sonner `toast`; send the `configOf` JSON that Copy and Download write to your server there
- The drawing lives in `SCENE` in server-scene.ts, sorted once by `paintOrder`; each model lists `gpus` that must match a card in `CARD_X`

## When not to use

- Planning next month's hosted model spend against a budget: use [Usage Planner](https://reui.io/blocks/ai-agents/ai-usage/ai-usage-1.md) (`get_block("ai-usage-1")`)

## Third-party libraries

- Sonner v2: [API reference](https://sonner.emilkowal.ski/toast), [docs](https://sonner.emilkowal.ski/getting-started)

## Install

```bash
npx shadcn@latest add @reui/ai-usage-4
```

## About AI Usage blocks

### When to use

- A self-serve planner where an admin sets next month's token volume, model and add-ons against a budget
- A month to date breakdown of AI spend by model, team, agent and API key, with the caps at risk
- Inference on your own hardware: edge nodes under a power budget, or open weight models sized to a GPU server

### When not to use

- Staff back office for invoices, subscriptions, dunning and metered charges: use the [Billing blocks](https://reui.io/blocks/solutions/billing.md)
- Public plan tiers or an upgrade paywall shown before anyone has usage: use the [Pricing blocks](https://reui.io/blocks/marketing/pricing.md)
- An operator console for provider health, routing and quotas across an org: use the [AI Ops blocks](https://reui.io/blocks/solutions/ai-ops.md)

### Agent notes

- `AiUsage` takes no props and calls no API; its figures are typed constants, mostly in components/data.tsx, so swap them for your data
- Money prints through en-US USD helpers such as `formatDollars` in formatters.ts, so another currency starts there

Every ReUI item has a twin like this one; the index is https://reui.io/llms.txt. Agents with the ReUI MCP (https://reui.io/docs/mcp) get the same guidance from the getters named above.